Use LM Studio for local AI in your notes

Thread can use a compatible model running on your own computer for titles, summaries, tags, meaning-based search and written answers. Model requests remain on that computer, and Orbit adds no AI charge; the software, model and hardware resources are yours.

Set up LM Studio

1. Download and install LM Studio from lmstudio.ai.
2. In Discover, download a chat model — e.g. qwen2.5-7b-instruct (fast, great for most laptops) or gemma 4B — and the embedding model text-embedding-nomic-embed-text-v1.5.
3. Open the Developer tab (server). Turn on Enable CORS — required so your browser may talk to LM Studio.
4. Start the server. It listens on http://localhost:1234.
5. Load your chat model (and the embedding model).
6. In Thread open Settings → AI, choose On this computer. Thread finds LM Studio automatically — pick the models for writing and for finding related notes, or leave them on Automatic.

Browser permission: local network access

Browsers now ask before a website may talk to apps on your computer. In Chrome, Edge or Brave, choose Allow when asked about local network access. Blocked it by accident? Click the icon left of the address bar → Site settings → Local network access → Allow, then reload. Safari doesn’t allow this at all — use Chrome, Edge, Brave or Firefox for local AI (your notes sync between browsers when you’re signed in), or use your own API key.

Ollama instead

Start Ollama with your site allowed as origin, e.g. OLLAMA_ORIGINS=https://your-thread-domain ollama serve, pull a model (ollama pull qwen2.5:7b) and select On this computer in Settings → AI.

Performance tips

• 7B–8B models can run well on recent Apple Silicon or GPUs with about 8 GB+; speed depends on the model and hardware.
• AI work happens in the background; capture and lexical search do not require it.
• If the model is unavailable, Thread reports the failure and you can retry later; saved source content remains available.